RECOMMENDED PRACTICE

DNV-RP-C207

STATISTICAL REPRESENTATION
OF SOIL DATA
OCTOBER 2010

DET NORSKE VERITAS

FOREWORD
DET NORSKE VERITAS (DNV) is an autonomous and independent foundation with the objectives of safeguarding life,
property and the environment, at sea and onshore. DNV undertakes classification, certification, and other verification and
consultancy services relating to quality of ships, offshore units and installations, and onshore industries worldwide, and carries
out research in relation to these functions.
DNV service documents consist of amongst other the following types of documents:
— Service Specifications. Procedual requirements.
— Standards. Technical requirements.
— Recommended Practices. Guidance.
The Standards and Recommended Practices are offered within the following areas:
A) Qualification, Quality and Safety Methodology
B) Materials Technology
C) Structures
D) Systems
E) Special Facilities
F) Pipelines and Risers
G) Asset Operation
H) Marine Operations
J) Cleaner Energy
O) Subsea Systems

The electronic pdf version of this document found through http://www.dnv.com is the officially binding version
© Det Norske Veritas

Any comments may be sent by e-mail to rules@dnv.com

For subscription orders or information about subscription terms, please use distribution@dnv.com
Computer Typesetting (Adobe Frame Maker) by Det Norske Veritas

If any person suffers loss or damage which is proved to have been caused by any negligent act or omission of Det Norske Veritas, then Det Norske Veritas shall pay compensation to such person
for his proved direct loss or damage. However, the compensation shall not exceed an amount equal to ten times the fee charged for the service in question, provided that the maximum compen-
sation shall never exceed USD 2 million.
In this provision "Det Norske Veritas" shall mean the Foundation Det Norske Veritas as well as all its subsidiaries, directors, officers, employees, agents and any other acting on behalf of Det
Norske Veritas.
 Recommended Practice DNV-RP-C207, October 2010
Changes – Page 3

CHANGES
• General • Main changes
As of October 2010 all DNV service documents are primarily Since the previous edition (April 2007), this document has
published electronically. been amended, most recently in October 2007. All changes
have been incorporated and a new date (October 2010) has
In order to ensure a practical transition from the “print” scheme been given as explained under “General”.
to the “electronic” scheme, all documents having incorporated
amendments and corrections more recent than the date of the
latest printed issue, have been given the date October 2010.
An overview of DNV service documents, their update status
and historical “amendments and corrections” may be found
through http://www.dnv.com/resources/rules_standards/.

DET NORSKE VERITAS

Recommended Practice DNV-RP-C207, October 2010
Page 4 – Changes

DET NORSKE VERITAS

 Recommended Practice DNV-RP-C207, October 2010
Page 5

CONTENTS

1. INTRODUCTION .................................................. 7 2.4.4 Calculation example for dependent soil variable ............. 17
2.4.5 Outliers ............................................................................. 17
1.1 General .....................................................................7 2.4.6 Data from more than one source....................................... 17
1.1.1 General................................................................................ 7 2.4.7 Bayesian approach to estimation ...................................... 17
1.1.2 Objective............................................................................. 7
1.1.3 Scope and application ......................................................... 7 2.5 Estimation of soil parameters with confidence... 17
2.5.1 Parameter estimation with confidence, independent
1.2 References ................................................................7 variables............................................................................ 17
1.2.1 Offshore Standards ............................................................. 7 2.5.2 Parameter estimation with confidence,
1.2.2 Recommended Practices and Classification Notes............. 7 dependent variables ......................................................... 18
1.3 Definitions ................................................................7 2.5.3 Calculation example for independent soil variable .......... 19
1.3.1 Verbal forms ....................................................................... 7 2.5.4 Calculation example for dependent soil variable ............. 19
1.3.2 Terms .................................................................................. 7 2.5.5 Minimum sample number................................................. 19
2.5.6 Bayesian approach to estimation with confidence ........... 20
1.4 Abbreviations and symbols.....................................8
1.4.1 Abbreviations...................................................................... 8 2.6 Combination of variances .................................... 20
1.4.2 Symbols .............................................................................. 8 2.6.1 Combination rules ............................................................ 20
2.6.2 Calculation example for independent soil variable .......... 20
2. STATISTICAL REPRESENTATION
OF SOIL DATA...................................................... 8 2.7 Geostatistics ........................................................... 21
2.7.1 Spatial averaging .............................................................. 21
2.1 Introduction .............................................................8 2.7.2 Example of spatial averaging ........................................... 21
2.1.1 General................................................................................ 8 2.7.3 Spatial estimation – kriging ............................................. 21
2.2 Types of uncertainty................................................9 2.7.4 Calculation example of kriging ....................................... 22
2.2.1 General ............................................................................... 9 2.8 Full probability distribution representation
2.2.2 Aleatory uncertainty .......................................................... 9 of soil properties with examples of application .. 23
2.2.3 Epistemic uncertainty ........................................................ 9 2.8.1 General.............................................................................. 23
2.2.4 Other types of uncertainty ................................................. 9 2.8.2 Application to geotechnical reliability analyses............... 23
2.2.5 Correlation of uncertainties ................................................ 9
2.9 Design soil parameters.......................................... 24
2.3 Uncertainty modelling.............................................9 2.9.1 General.............................................................................. 24
2.3.1 Probability distributions .................................................... 9 2.9.2 Characteristic value .......................................................... 24
2.3.2 Definitions, symbols and notions for probability 2.9.3 Best estimate..................................................................... 24
distributions ...................................................................... 10
2.3.3 The normal distribution ................................................... 13 2.9.4 Lower and upper bounds .................................................. 25
2.3.4 Other distributions ............................................................ 13 2.10 References and literature ..................................... 25
2.3.5 Characteristic values......................................................... 14 2.10.1 List of references .............................................................. 25
2.4 Central estimation of soil parameters .................14 2.10.2 Other relevant literature.................................................... 25
2.4.1 Parameter estimation for independent soil variables ........ 14
2.4.2 Parameter estimation for dependent soil variables ........... 15 APP. A TABLES OF PROBABILITY
2.4.3 Calculation example for independent soil variable .......... 16 DISTRIBUTIONS ............................................................. 26

DET NORSKE VERITAS

Recommended Practice DNV-RP-C207, October 2010
Page 6

DET NORSKE VERITAS

 Recommended Practice DNV-RP-C207, October 2010
Page 7

1. Introduction 1.2 References

1.1 General 1.2.1 Offshore Standards
1.1.1 General 1.2.1.1 The latest versions of the documents in Table 1-1 apply.
1.1.1.1 This Recommended Practice provides principles, Table 1-1 DNV Offshore Standards
guidance and recommendations for use of statistical methods
for analysis and representation of soil data. Reference Title
DNV-OS-C101 Design of Offshore Steel Structures,
1.1.1.2 Soil data to be used as a basis for geotechnical analysis General (LRFD Method)
and foundation design are usually encumbered with uncer- DNV-OS-F101 Submarine Pipeline Systems
tainty, both in terms of natural variability and in terms of lim-
ited amounts of data. Statistical methods form a rational tool DNV-OS-F201 Dynamic Risers
for handling of such uncertain data, and this Recommended DNV-OS-J101 Design of Offshore Wind Turbine
Practice shows how to apply this tool, in particular with a view Structures
to the selection of characteristic values for use in design.
1.2.2 Recommended Practices and Classification Notes
1.1.1.3 The selection of characteristic values of soil properties
for use in geotechnical design is often based on subjective 1.2.2.1 The Recommended Practices and Classification Notes
judgment and accumulated experience, and the uncertainties in Table 1-2 form useful references for application of this doc-
which are involved with the soil properties are only to a limited ument.
extent brought into the picture when the characteristic values
are chosen. Table 1-2 DNV Recommended Practices and Classification
Notes
1.1.2 Objective
Reference Title
1.1.2.1 It is a major purpose of this Recommended Practice to Classification Notes 30.4 Foundations
demonstrate how characteristic values of soil properties can be Classification Notes 30.6 Structural Reliability of Marine
extracted from available data in a rational manner by means of Structures
statistical methods. Statistical methods allow for quantifica-
tion of the involved uncertainties and make it possible to DNV-RP-E301 Design and Installation of Fluke
account for these uncertainties in a rational manner. The focus Anchors in Clay
is on application of statistical methods to rational selection of DNV-RP-E302 Design and Installation of Plate
characteristic values for design, but other applications of sta- Anchors in Clay
tistical methods are also considered, e.g. development of full DNV-RP-E303 Geotechnical Design and
probabilistic models of soil properties for use in probabilistic Installation of Suction Anchors
analyses such as structural reliability analyses. in Clay
1.1.3 Scope and application DNV-RP-F105 Free Spanning Pipelines

1.1.3.1 The Recommended Practice is applicable to analysis 1.3 Definitions

and representation of soil data for all kinds of offshore geo-
technical engineering problems.
1.1.3.2 The following major topics are covered by the Recom-
mended Practice:
— types of uncertainties in soil data
— uncertainty modelling, including probability distributions
and parameters to describe them
— generic distribution types for soil data
— central estimation of soil parameters
— estimation of soil parameters with confidence
— interpretation of characteristic values from data
— geostatistics, including spatial averaging and spatial esti-
mation
— full probability distribution modelling of soil properties in
geotechnical reliability analysis.
1.3.1 Verbal forms
1.3.1.1 Shall: Indicates a mandatory requirement to be fol-
lowed for fulfilment or compliance with the present Recom-
mended Practice. Deviations are not permitted unless formally
and rigorously justified, and accepted by all relevant contract-
ing parties.
1.3.1.2 Should: Indicates a recommendation that a certain
course of action is preferred or is particularly suitable. Alterna-
tive courses of action are allowable under the Recommended
Practice where agreed between contracting parties, but shall be
justified and documented.
1.3.1.3 May: Indicates permission, or an option, which is per-
mitted as part of conformance with the Recommended Prac-
tice.
1.3.1.4 Can: Requirements with can are conditional and indi-
cate a possibility to the user of the Recommended Practice.
1.3.1.5 Agreement, or by agreement: Unless otherwise indi-
cated, agreed in writing between contractor and purchaser.
1.3.2 Terms
1.3.2.1 Characteristic value: A representative value of a load
variable or a resistance variable. For a load variable, it is a high
but measurable value with a prescribed probability of not being
unfavourably exceeded during some reference period. For a
resistance variable it is a low but measurable value with a pre-
scribed probability of being favourably exceeded.

DET NORSKE VERITAS

Recommended Practice DNV-RP-C207, October 2010
Page 8

1.3.2.2 Classification Notes: The Classification Notes cover s Position, location coordinates
proven technology and solutions which are found to represent
good practice by DNV, and which represent one alternative for se Standard error; standard deviation of estimate
satisfying the requirements stipulated in the DNV Rules or su Undrained shear strength
other codes and standards cited by DNV. The classification suC Static triaxial compression undrained shear strength
notes will in the same manner be applicable for fulfilling the
requirements in the DNV offshore standards. suD Static DSS undrained shear strength
suE Static triaxial extension undrained shear strength
1.4 Abbreviations and symbols t Parameter in Beta distribution
1.4.1 Abbreviations tn Quantile in Student’s t distribution
v Generic variable in expression for Beta distribution
1.4.1.1 Abbreviations listed in Table 1-3 are used in this Rec- w Weighting factor in regression analysis
ommended Practice.
x Realisation of stochastic variable
Table 1-3 Abbreviations X Stochastic variable, random variable
Abbreviation In full X Vector of stochastic variables
CIU Consolidated isotropic undrained (test) y Realisation of stochastic variable
CPT Cone penetration test Y Stochastic variable, random variable
DSS Direct simple shear z Depth
FORM First-order reliability method Z Stochastic variable, random variable
SORM Second-order reliability method 1.4.2.2 Greek characters
UU Unconsolidated undrained (test)
 Dimensionless factor for axial pile capacity
1.4.2 Symbols
 Scale parameter in Weibull distribution
1.4.2.1 Latin characters  Complement of confidence 1
3 Skewness
a Scale parameter
4 Kurtosis
a Lower bound for bounded variable
 Shape parameter in Weibull distribution
a0 Soil surface intercept
 Reliability index
a1 Depth gradient
 Residual, difference between observed value and
b Limit of autocorrelation function for small lag r predicted value
b Upper bound for bounded variables  Standard normal cumulative distribution function
C Covariance, covariance function  Variogram function
C Covariance matrix  Shift parameter in Weibull distribution
Cov Covariance  Variance reduction factor to account for effect of spatial
COV Coefficient of variation averaging
c1 Quantile in non-central t distribution  Weighting factor in point estimation by kriging
D Diameter  Mean value
E Mean value operator  Coefficient of correlation, autocorrelation function
f Probability density function  Correlation matrix
F Cumulative distribution function  Standard deviation
g Limit state function
k Proportionality factor between standard deviation and
depth 2. Statistical Representation of Soil Data
L Length
L Load 2.1 Introduction
m Mode in probability distribution 2.1.1 General
n Sample size, number of data
n Degrees of freedom 2.1.1.1 Following an introduction to uncertainty types in soil
data and to principles of uncertainty modelling, this section
p Probability, fraction presents statistical methods for analysis and representation of
P Probability soil data. Formulas for central estimation of soil parameters
PF Probability of failure and formulas for estimation of soil parameters with confidence
Q Probability of exceedance, complement of cumulative are given. Examples of application are given with focus on
distribution function estimation of characteristic values. Guidance for selection of
soil parameters for design is given.
Q Capacity
r Parameter in Beta distribution 2.1.1.2 Issues of special interest in the context of uncertainty
r Lag, distance, separation modelling for soil data for large volumes, such as spatial aver-
aging and spatial estimation, are dealt with in separate subsec-
R Correlation length tions, and examples of application are given.
R Resistance
s Estimate of standard deviation

DET NORSKE VERITAS


2.1.1.3 An introduction to full probabilistic modelling of soil
properties is given.
2.2 Types of uncertainty
2.2.1 General
2.2.1.1 Uncertainties associated with geotechnical data have
many sources and may be divided into the following two main
types of uncertainty:
— aleatory uncertainty, i.e. physical uncertainty
— epistemic uncertainty, i.e. uncertainty related to imperfect
knowledge.
2.2.2 Aleatory uncertainty
2.2.2.1 Aleatory uncertainty consists of physical uncertainty.
Physical uncertainty is also known as inherent uncertainty and
intrinsic uncertainty and is a natural randomness of a quantity
such as the variability in the soil strength from point to point
within a soil volume. Such physical uncertainty or natural var-
iability is a type of uncertainty which cannot be reduced.
2.2.2.2 Physical uncertainty associated with a geotechnical
variable is a fundamental property of the variable.
2.2.3 Epistemic uncertainty
2.2.3.1 Epistemic uncertainty consists of statistical uncer-
tainty, model uncertainty and measurement uncertainty, which
are all classified as a type of uncertainty associated with lim-
ited, insufficient or imprecise knowledge.
2.2.3.2 Epistemic uncertainty can in principle be reduced by
collection of more data, by improving engineering models and
by employing more accurate methods of measurement.
2.2.3.3 Statistical uncertainty is uncertainty due to limited
information such as a limited number of observations of a
quantity, e.g. a limited number of soil strength values from a
limited number of soil samples tested in the laboratory.
2.2.3.4 Model uncertainty is uncertainty due to imperfections
and idealisations made in (1) applied engineering models for
representation and prediction of quantities such as pile resist-
ance and foundation capacity and in (2) choices of probability
distribution types for representation of uncertain quantities.
2.2.3.5 Model uncertainty involves two elements, viz. (1) a
bias if the model systematically leads to overprediction or
underprediction of a quantity in question and (2) a randomness
associated with the variability in the predictions from one pre-
diction of that quantity to another.
Guidance note:
A geotechnical example of model uncertainty is the uncertainty
associated with axial pile capacity predictions from undrained
shear strength by the relatively simple  method.
Another example of model uncertainty is the uncertainty associ-
ated with the conversion of cone resistance data from CPT tests
to undrained shear strength data.
A third example of model uncertainty is the uncertainty associ-
ated with the choice of a generic distribution type for the proba-
bility distribution of soil properties when data are too scarce to
identify the correct distribution type, when no generic distribu-
tion type provides a good fit to data, or when a wrong distribution
type is used in calculations for simplicity.
---e-n-d---of---G-u-i-d-a-n-c-e---n-o-t-e---
2.2.3.6 Model uncertainty varies from case to case and can be
studied by interpretation of pertinent sensitivities.
2.2.3.7 Measurement uncertainty is caused by imperfect
instruments and sample disturbance when a quantity is
observed. Measurement uncertainty involves two elements,
DET NORSKE VERITAS
Recommended Practice DNV-RP-C207, October 2010
Page 9
viz. (1) a systematic error associated with the bias in the meas-
urements and (2) a random error associated with the precision
in the measurements.
Guidance note:
An example of measurement uncertainty, which may involve
both a bias and a random error, is the uncertainty associated with
sample disturbance of soil samples collected for subsequent test-
ing in the laboratory, e.g. for determination of the soil strength.
When sample disturbance takes place, the expected tendency will
be that a smaller strength than the true strength will be measured
in the laboratory, i.e. the sample disturbance implies introduction
of a systematic error in terms of a negative bias. If, in addition,
the degree of sample disturbance varies from sample to sample,
then a random error component in the measured strength is also
introduced.
---e-n-d---of---G-u-i-d-a-n-c-e---n-o-t-e---
2.2.3.8 If a considered quantity is not observed directly from
measurements, but predicted by imposing some data process-
ing of the observed data, a combined model and measurement
uncertainty may result.
Guidance note:
Interpretation of undrained shear strength from cone penetration
test results is an example of a procedure that leads to soil strength
data which are encumbered with a combined model and measure-
ment uncertainty.
---e-n-d---of---G-u-i-d-a-n-c-e---n-o-t-e---
2.2.3.9 Bias is defined as the difference between the expected
value of the prediction of a quantity by measurements, predic-
tion models or other means of estimation, and the true value of
the quantity.
2.2.4 Other types of uncertainty
2.2.4.1 Other types of uncertainties exist such as workmanship,
human errors and gross errors, but they are not considered here.
Guidance note:
For artificial soil deposits such as a backfill over a pipeline in a
trench, the uncertainty associated with the workmanship applied
for the installation of the backfill may appear in the same manner
as a natural variability. However, for any new backfills to be
installed this uncertainty can in principle be reduced by changing
and improving the process of work, if this is an option with the
available equipment.
---e-n-d---of---G-u-i-d-a-n-c-e---n-o-t-e---
2.2.5 Correlation of uncertainties
2.2.5.1 Some types of uncertainties may be correlated. Other
types of uncertainties are uncorrelated or independent. For
example, epistemic uncertainties are by nature statistically
independent of natural variability.
2.3 Uncertainty modelling
2.3.1 Probability distributions
2.3.1.1 All uncertainty types presented in 2.2.1 - 2.2.3 can in
principle be represented by a probability distribution. When a
parametric distribution model is used, the distribution can be
represented by its generic distribution type and its distribution
parameters.
2.3.1.2 In the context of soil properties, the most commonly
encountered generic distribution models consist of:
— normal distribution
— lognormal distribution
— weibull distribution
— uniform distribution
— triangular distribution.
Recommended Practice DNV-RP-C207, October 2010
Page 10

Guidance note: The probability density function is only defined for continu-
The normal distribution is typically used for variables which ous, differentiable cumulative distribution functions.
come about as sums or averages of a large number of independ-
ent quantities. Strength variables often tend to follow the normal The p quantile in the distribution of X is the value of X whose
distribution. The lognormal distribution is used for variables cumulative distribution function value is p, hence
1
which come about as non-negative products of a large number of x p  F X ( p)
independent quantities. The uniform distribution often forms a
suitable distribution model for physically bounded variables for
which no prior information exists. 2.3.2.3 The mean value of X is defined as

---e-n-d---of---G-u-i-d-a-n-c-e---n-o-t-e---
  E X    xf X ( x)dx

2.3.1.3 The most commonly encountered parameters used to and is also known as the expected value of X.
characterise a probability distribution consist of the following
properties: 2.3.2.4 The variance of X is


VarX    ( x   ) 2 f X ( x)dx
— mean value (expected value)
— standard deviation (a measure of the spread or the scatter 
about the mean)
— coefficient of variation (standard deviation divided by the which alternatively can be expressed as
mean value; a relative measure of the spread or the scatter Var X   E ( X   ) 2   E X 2   E X 
2

about the mean)

— lower bound and upper bound (used in the case of bounded and the standard deviation of X is
  Var X 
variables).
Guidance note:
The coefficient of variation is defined as
For some generic distribution types, such as the normal distribu-
tion, these parameters are used directly as distribution parameters 
COV 
in the mathematical expressions for the distributions. For other
generic distribution types, such as the log-normal distribution

and the Weibull distribution, distribution parameters are used Guidance note:
which can be derived from the listed characterising properties.
The coefficient of variation expresses the standard deviation  as
---e-n-d---of---G-u-i-d-a-n-c-e---n-o-t-e--- a fraction of the mean value . The coefficient of variation is
meaningful only for variables which have positive mean values.
For geotechnical applications the coefficient of variation is of
2.3.1.4 Other terms and properties which are often used in the practical use for variables whose realisations are non-negative,
characterisation of probability distributions include: such as soil strengths.
— variance, defined as the square of the standard deviation ---e-n-d---of---G-u-i-d-a-n-c-e---n-o-t-e---
— skewness, a measure of the asymmetry of the distribution
— kurtosis, a measure of the flatness of the distribution 2.3.2.5 The skewness of X is
— standard error, term used for the standard deviation of esti- 

 (x  )
3
mators for constant quantities, such as estimators for dis- f X ( x)dx
tribution parameters like the mean value 3  

— covariance, a measure of the joint variability of two quan- 3

tities or variables Guidance note:
— correlation, a relative measure of the joint variability of
Symmetric distributions such as the normal distribution have
two quantities or variables. 3 = 0. Distributions skewed to the right have 3 > 0. Distribu-
tions skewed to the left have 3 < 0. Examples are given in
2.3.1.5 Examples of parameters listed in 2.3.1.3 and 2.3.1.4 Figure 2-1.
can be found in 2.3.2.1 through 2.3.2.8.
---e-n-d---of---G-u-i-d-a-n-c-e---n-o-t-e---
2.3.2 Definitions, symbols and notions for probability dis-
tributions
2.3.2.1 Let X denote a random soil variable such as the
strength in a soil deposit, and let x be a realisation of X such as
the observed value of X in a particular point within the soil
deposit.
2.3.2.2 The cumulative distribution function FX of X is
defined as the probability that X falls short of x,
FX ( x )  P X  x 
where P[] denotes probability. Figure 2-1
The probability of exceedance QX is defined as the comple- Probability density functions to illustrate skewness
ment of the cumulative distribution function
Q X ( x )  1  F X ( x )  P X  x  2.3.2.6 The kurtosis of X is
The probability density function is defined as the derivative of 
the cumulative distribution function
 (x  )
4
f X ( x)dx
dF ( x)
f X ( x)  4  

dx 4

DET NORSKE VERITAS


Guidance note:
The normal distribution has 4 = 3. Long-tailed distributions
have 4 > 3. Flat-topped distributions which have a large concen-
tration of outcomes near the mean have 4 < 3. Examples are
given in Figure 2-2.
---e-n-d---of---G-u-i-d-a-n-c-e---n-o-t-e---
Figure 2-2
Probability density functions to illustrate kurtosis
2.3.2.7 The median of X is the 50% quantile in the distribution
of X. For a symmetrical distribution, such as the normal distri-
bution, the median of X coincides with the mean value of X.
2.3.2.8 Let X and Y denote two soil variables. The covariance
between X and Y is defined as
CovX , Y   E ( X   X )(Y   Y )
The covariance matrix for X and Y is defined as
 Var X  CovX , Y 
C XY   
CovX , Y  Var Y  
The coefficient of correlation between X and Y is defined as
CovX , Y 

Var X Var Y 
and is a relative measure of correlation which assumes values
between 1 and 1. A value of 0 indicates no correlation, a value
of 1 indicates full negative correlation, and a value of 1 indi-
cates full positive correlation. See Figure 2-3.
The correlation matrix for X and Y is defined as
1 
ρ XY    environments. This variation can exist even in apparently homo-
 1 
Guidance note:
The coefficient of correlation is a measure of closeness of linear
relationship between two variables X and Y. A positive value
indicates a tendency of X and Y to increase together. A negative
value indicates that large values of X are associated with small
values of Y. Values of either 1 or 1, corresponding to full posi-
tive and negative correlation, respectively, imply linear func-
tional relationships between X and Y. The diagram in the upper
left corner of Figure 2-3 visualises the situation with full negative
correlation between X and Y and the associated linear relation-
ship.
The diagram in the lower right corner of Figure 2-3 visualises
that whereas there may be a strong dependency between two var-
iables X and Y, this does not always imply a strong correlation
between the variables. For example, Y = X2 is very much
dependent on X, but the coefficient of correlation between X and
Y is 0.
In general, if X and Y are independent, the coefficient of corre-
lation between them is 0. However, the converse is not necessar-
ily true.
---e-n-d---of---G-u-i-d-a-n-c-e---n-o-t-e---
DET NORSKE VERITAS
Recommended Practice DNV-RP-C207, October 2010
Page 11
Figure 2-3
Examples of coefficients of correlation
2.3.2.9 A correlation matrix can also be established for the
elements of an array of three or more variables. For an array of
three variables, Z = (Z1,Z2,Z3), the correlation matrix has the
form
 1  12  13 
ρ Z    12 1  23 
  13  23 1 
Guidance note:
An example of an array of three correlated variables that are
commonly encountered in geotechnical problems is the array of
undrained shear strengths in compression, direct simple shear
and extension, (suC, suD, suE).
---e-n-d---of---G-u-i-d-a-n-c-e---n-o-t-e---
2.3.2.10 Let X denote a soil variable such as the soil strength
within a soil volume. X exhibits natural variability and there-
fore varies from point to point within the soil volume. X is said
to form a random field over the soil volume.
Guidance note:
The natural variability from point to point within a soil volume is
a result of the natural formation of soil in different depositional
geneous soil units.
---e-n-d---of---G-u-i-d-a-n-c-e---n-o-t-e---
2.3.2.11 The autocorrelation function for X is defined as the
coefficient of correlation between the values of X in two points
located a distance r apart within the soil volume,
CovX (s), X (s' )
 (r ) 
Var X 
where s and s’ are the location coordinates of the two points
and r = |ss’| denotes the distance between them.
The autocorrelation function reflects the connectivity in the
soil properties and tells something about how the variation of
X from point to point is. An autocorrelation function typically
decreases for increasing lag r.
Guidance note:
A commonly applied model for the autocorrelation function for
soil properties is the quadratic exponential decay model,
r2
 (r )  exp( )
R2
Recommended Practice DNV-RP-C207, October 2010
Page 12

in which R is a correlation length which expresses the scale of Figure 2-5 shows a variogram with nugget effect, range and
fluctuation in the random field X. Random fields of soil proper- sill. The nugget effect is defined in 2.3.2.12. The range is a dis-
ties are usually anisotropic. The vertical correlation length RV is tance beyond which the variogram essentially remains con-
usually smaller than the horizontal correlation length RH. This stant. The sill is the plateau the variogram reaches at the range.
difference reflects the geological processes that lead to the for-
mation of soil deposits by sedimentary depositing and may
amount to as much as an order of magnitude.
Let r have components x in the horizontal plane and z in the
vertical direction. For the anisotropic case that RV  RH, the
quadratic exponential decay model can then be expressed as
x 2 z
 (x, z )  exp(( )  ( )2 )
RH RV
Examples of autocorrelation function models are given in
Figure 2-4.
---e-n-d---of---G-u-i-d-a-n-c-e---n-o-t-e---

0.8
Figure 2-5
Variogram
0.6
(r)
0.4
Guidance note:
0.2 The variogram (r) is slightly more flexible to operate with than
the covariance function C(r), because there are more licit vario-
0 gram functions than there are licit covariance functions. This is
in the sense that licit variogram functions, which do not have a
0 0.5 1 1.5 2 2.5 3 sill for large lags r do not have a covariance function counterpart.
r/R The variogram (r) is sometimes denoted semivariogram, which
refers to its alternative expression as (r) = ½Var[X(s)–X(s’)].
Quadr. exponential Exponential The most commonly used variogram models with sill are

Figure 2-4 3 r 1 r

 (r )   2 R  2 ( R ) rR
3
Autocorrelation function models Spherical model:
 1 rR
2.3.2.12 The theoretical limit of the autocorrelation function r
as the lag r approaches zero is unity, Exponential model:  (r )  1  exp( )
R
(r)  1 for r  0.
Quadratic exponential r 2
However, when (r) is inferred from data, one may observe decay model:  (r )  1  exp(( ) )
that R
(r)  b for r  0 with 0 < b < 1 Hole effect model:  (r )  1 
sin( ar )
The difference 1b is known as “nugget effect”. The nugget r
effect is in most cases due to measurement uncertainty, which The most commonly used variogram models without sill are
by nature does not exhibit any spatial correlation structure.
Guidance note:
 (r )  r ; 0  a  2
Power model: a
The presence of a nugget effect will imply that the quadratic
exponential decay model becomes changed to Hole effect model:  (r )  1  cos(ar )
2
r
 (r )  b  exp( ) for r  0 The coefficients a and R are scale parameters, and R is sometimes
R2 referred to as correlation length.
---e-n-d---of---G-u-i-d-a-n-c-e---n-o-t-e---
---e-n-d---of---G-u-i-d-a-n-c-e---n-o-t-e---
2.3.2.14 Autocorrelation functions and variograms can in
2.3.2.13 The variogram associated with the random field X is principle be estimated from data, provided a sufficient amount
an alternative measure of the spatial connectivity of X and is of data for the relevant parameter is available from the soil vol-
defined as ume of interest. Autocorrelation function models and vario-
 (r )  C (0)  C (r )   2  (1   (r )) gram models can then be fitted to the estimated functions. The
model that provides the best fit to the estimated function values
in which should be sought after. In practice, the amount of data will
C (r )  CovX (s), X (s' )   2   (r ); r | s  s' | often be scarce, in which case one may be left with choosing a
model on the basis of experience. Also in practice, one may
is the covariance function with the special case sometimes find that it is of less importance which model is
C (0)  Var X    2 used as long as the scale parameter is correct. It may facilitate
the estimation of autocorrelation functions and variograms
where s and s’ denote the location coordinates of two points from data to sample the data equidistantly, provided such equi-
located a distance r = |ss’| apart. distant sampling is feasible.

DET NORSKE VERITAS

 Recommended Practice DNV-RP-C207, October 2010
Page 13

2.3.2.15 Autocorrelation functions and variograms are used
for estimation of soil properties and associated uncertainties in
problems that involve large areas and soil volumes. For exam-
ples of application, reference is made to 2.7.1 - 2.7.4.
2.3.3 The normal distribution
2.3.3.1 The normal probability distribution is a distribution
model which often proves adequate for representation of a
number of soil properties, including strength properties such as
the undrained shear strength. The normal distribution is fully
described by its mean value and its standard deviation, i.e. two
parameters are all that is required for a complete description of
this distribution model.
2.3.3.2 The cumulative distribution function for a considered
soil variable X gives the cumulative probability up to a speci-
fied outcome x of X, i.e. the probability that X will fall short of
x. For a normally distributed variable X, the cumulative distri-
bution function reads
x
FX ( x )  P X  x    ( ) Note that the natural logarithm of the mean value of X is not the
 mean value of lnX.
1 x
(x  )2
 exp(
 )dx
 2   2
2.3.4 Other distributions
2.3.4.1 The lognormal distribution is defined such that if the
variable X is lognormally distributed, then the variable lnX is
normally distributed. The lognormal distribution is only
defined for positive outcomes of X and is therefore sometimes
used to represent distributions of variables that cannot take on
negative values, such as strengths. A drawback by using the
lognormal distribution for strength variables is that it may
sometimes lead to nonconservative predictions of strengths in
the lower tail of the distribution.
Guidance note:
The following formulas can be used to transform the mean value
and the variance of lnX to mean value and variance of X, when X
is lognormally distributed
E X   exp( E ln X   Var ln X )
1
2
Var X   E X   exp(Var ln X   1)
2
---e-n-d---of---G-u-i-d-a-n-c-e---n-o-t-e---

2.3.4.2 The Weibull distribution has a cumulative distribution

in which  denotes the standard normal cumulative distribu- function
tion function,  denotes the mean value of X and  denotes the
x  
standard deviation of X. A table of the standard cumulative FX ( x )  1  exp(( ) )
normal distribution is given in Appendix A. 
in which the scale parameter , the shape parameter  and the
2.3.3.3 For the situation that there is only a limited number of
laboratory test data on soil samples available for the determi- shift parameter  are distribution parameters which all relate to
nation of the distribution of X, the estimates of the two distri- the mean value and standard deviation of the distribution. The
bution parameters (the mean value and the standard deviation) Weibull distribution is a flexible distribution which has many
from the data will be encumbered with statistical uncertainty, applications, e.g. distributions of local maxima of properties
i.e. the mean value and the standard deviation will themselves that vary in time and space. The exponential distribution and
have uncertainties that can be modelled by assigning probabil- the Rayleigh distribution are special cases of the Weibull dis-
ity distributions to them. These distributions will have the cal- tribution for = 1 and = 2, respectively.
culated estimates as their mean values and they will in addition
be characterised by their standard errors. 2.3.4.3 The Beta distribution is a flexible distribution model
for variables which are upwards and downwards bounded and
2.3.3.4 Variables that are modelled to follow the normal dis- for which the mean value and the standard deviation are
tribution can take on negative values. This may be unphysical known.
for a number of soil variables such as soil strengths, which For a variable X which is Beta-distributed over the range [a; b]
never take on negative values, but for which the normal distri- and has mean value  and standard deviation , the cumulative
bution still is an adequate distribution model, e.g. as judged on distribution function reads
other physical grounds than the extent of the physically possi- x
ble space of realisations. Usually, it will not pose any problems
 (v  a )
r 1
that the normal distribution in theory can take on negative val- (b  v)t  r 1 dv
ues, since for practical purposes the probability of negative FX ( x)  a
; a  x  b and 0  r  t
values will usually be negligible. Only in cases where the coef- (b  a)t 1 B(r , t  r )
ficient of variation of a variable in question is particularly where B denotes the Beta function and in which the parameters
large, there may be a risk of getting negative outcomes of the r and t relate to  and  through
variable when a normal distribution model is used. In such
cases it is recommended to apply a truncated normal distribu- r r tr
tion, where the point of truncation of the lower tail is set at   a  (b  a ) and   (b  a )
t t r (t  1)
zero.
Guidance note: 2.3.4.4 The uniform distribution is a useful distribution model
Let X denote a normal distribution with mean value  and stand- for variables for which physical upper and lower bounds are
ard deviation , and let  denote the standard normal cumulative known while no other prior information exists.
distribution function. Let Y be generated as X truncated at x = 0.
The cumulative distribution function for Y is then defined as For a variable X which is uniformly distributed over the range
y  [a; b], the cumulative distribution function reads
( )   ( ) xa
FY ( y )    ; y0 FX ( x) 
 ba
1   ( )
 The mean value is
ab
---e-n-d---of---G-u-i-d-a-n-c-e---n-o-t-e--- E X  
2

DET NORSKE VERITAS

Recommended Practice DNV-RP-C207, October 2010
Page 14
and the standard deviation is
ba
X 
12
2.3.4.5 The triangular distribution is a useful distribution
model for variables for which physical upper and lower bounds
are known together with the most likely value.
For a variable X which follows a triangular distribution over
the range [a; b] and whose most likely value is m, the cumula-
tive distribution function reads
 ( x  a) 2
 axm
(m  a )(b  a )
FX ( x)  
1  (b  x )
2
m xb
 (b  m)(b  a)
The mean value is
abm
E X  
3 literally interpreted as the smallest value among n observed val-
and the standard deviation is
a 2  b 2  m 2  ab  am  bm
X 
18
2.3.5 Characteristic values
2.3.5.1 For practical purposes in deterministic geotechnical
design, one will not apply the entire probability distribution of
a quantity in question such as the soil strength, but rather a
characteristic value which is then usually defined as a particu-
lar quantity associated with the probability distribution. Exam-
ples of such quantities which are often used as characteristic
values are:
— mean value
— a low quantile in the probability distribution, e.g. the 5%
quantile
— mean value minus two standard deviations, which for a
normal distribution corresponds to the 2.3% quantile
— most probable value, i.e. the value for which the probabil-
ity density function is maximum.
Guidance note:
Which definition of the characteristic value applies usually
depends on the design code and on the actual application.
For problems which are governed by a local soil strength value,
such as the pile tip resistance for an end-bearing pile, a low quan-
tile in the strength distribution is often specified as characteristic
value.
For problems, which involve large soil volumes where local
strength variations from point to point can be assumed to average
out, such as in stability calculations for large gravity-base foun-
dations, the mean value of the soil strength is often specified as
characteristic value.
---e-n-d---of---G-u-i-d-a-n-c-e---n-o-t-e---
2.3.5.2 For the case that the distribution of the considered soil
strength is estimated based on limited test data such that the
distribution parameters become encumbered with statistical
uncertainty, also estimates of characteristic values of the distri-
bution become statistically uncertain. To properly account for
such statistical uncertainty, it is common to specify that char-
acteristic values shall be estimated with a specified confidence.
2.3.5.3 Estimation with confidence implies that instead of
using a central estimate of the quantity in question, a conserv-
ative estimate value is used such that the probability of the true
value being more favourable than the conservative estimate is
at least equal to the specified confidence.
DET NORSKE VERITAS
Guidance note:
When the characteristic value in a design code is defined as the
5% quantile in the probability distribution, then an additional
requirement usually is that it shall be estimated with a specified
confidence. A confidence of 95% is a typical requirement in
many codes.
Some design codes, which use the mean value as characteristic
value, specify that the characteristic value shall be taken as “a
conservatively assessed mean value” or “a cautiously estimated
mean value”. The interpretation of this specification is that the
characteristic value shall be taken as the mean value estimated
with a confidence greater than 50%. Typical confidences would
then be 75% or 90%.
---e-n-d---of---G-u-i-d-a-n-c-e---n-o-t-e---
2.3.5.4 Some geotechnical practices define the characteristic
value as a lower bound value with reference to some available
data set of n observations of the parameter in question.
Guidance note:
Note that a lower bound value is a subjective measure, and if it is
ues, a smaller value is likely to be obtained if the number of
observations n is increased. This is so, because the larger the
number of observations, the higher is the probability that one of
them will be made further out on the lower tail of the distribution.
Hence by using the observed lowest value as characteristic value,
one may actually loose the benefit of carrying out a larger
number of observations.
Note also that a lower bound by its mere definition is not neces-
sarily invariant with the number of observations n and does thus
not necessarily result in a specific quantile whose exceedance
probability is independent of n.
A lower bound value specified in a soil report to be used as char-
acteristic value for design is not necessarily a physical lower
bound according to the mathematical definition, but rather a rec-
ommended low value to be used in design.
It is therefore recommended that the designer always makes sure
that such specified lower bound values meet the requirements to
the characteristic value as set forth in the design codes and stand-
ards which are to be used in design.
When a characteristic value in terms of a so-called lower bound
is to be established and is to be presented in a soil report as a
lower bound for use in design, it is recommended that it be estab-
lished on the basis of statistical methods in accordance with the
examples of definitions given in 2.3.5.1. A mean value should be
used as characteristic value when the average of the soil property
in question governs the design, whereas a low quantile should be
used when a point value of the property governs the design. An
adequate confidence should be used for the estimation of the
characteristic value from data.
---e-n-d---of---G-u-i-d-a-n-c-e---n-o-t-e---
2.4 Central estimation of soil parameters
2.4.1 Parameter estimation for independent soil variables
2.4.1.1 Let X be a soil variable, such as a strength, for which
n observations exist, x1,...xn. The mean value of X is  and the
standard deviation is . Any dependency on depth or on any
other conditioning quantity is disregarded here; i.e. X is a so-
called independent variable.
2.4.1.2 The mean value  is estimated by the sample mean
1 n
x   xi
n i 1
2.4.1.3 The standard deviation  is estimated by the sample
standard deviation
1 n
s 
n  1 i 1
( xi  x ) 2
 Recommended Practice DNV-RP-C207, October 2010
Page 15

2.4.1.4 The estimates x and s2 are central estimates of 

and 2. a0 Y

2.4.1.5 The estimates x and s are encumbered with uncer- 

tainty owing to the limited number of data used for the estima- a1
tion. The uncertainties in the estimates can be represented by
the respective standard deviations, denoted standard errors. 1
If needed, these standard errors can be estimated by
s  1
se( x )  and se( s )  s 4 z
n 4n
Figure 2-6
There is also an expression for the standard error of the vari- Data points (z,Y) and linear model for mean
ance estimate s2,
2
se( s 2 )  s 2 2.4.2.2 The coefficients a0 and a1 in the linear expression can
n 1 be estimated from the n observed data pairs (z,y) as
n
2.4.1.6 The median xm of X is estimated by the sample
median. Let the n observations of X be arranged in increasing
 (z i  z )( yi  y )
order, x1,...xn. When n is odd, the sample median is the middle aˆ1  i 1
n
and aˆ0  x  aˆ1 z
value x(n+1)/2 in the ordered list of observations. When n is  ( zi  z ) 2
i 1
even, the sample median is the average of the two middle val-
ues xn/2 and x(n+2)/2 in this list. in which
Guidance note: 1 n 1 n
y 
n i 1
yi and z   zi
n i 1
For the special case that the distribution of X is known to be a
normal distribution, the median xm can be estimated by the sam-
Guidance note:
ple mean
The estimates â0 and â1 are known as best estimates of a0 and a1,
1 n
xˆ m   xi
n i 1
respectively, and thus provide the best estimate of the straight
line that expresses the mean variation of the variable Y with the
depth z.
hence utilising that the symmetry of the normal distribution
implies that the median is equal to the mean. ---e-n-d---of---G-u-i-d-a-n-c-e---n-o-t-e---
For the special case that the distribution of X is known to be a
lognormal distribution, the median xm can be estimated by the 2.4.2.3 The standard deviation  of the natural variability term
sample geometric mean  is estimated by the sample standard deviation
n
1 n
xˆ m  n x
i 1
i
s  ( yi  (aˆ 0  aˆ1 z i )) 2
n  2 i 1
Note that the lognormal distribution is not a symmetrical distri-
bution, such that the mean will be different from the median and 2.4.2.4 The estimators â0, â1 and s are encumbered with
such that the geometric mean of the observations used to estimate uncertainty owing to the limited number of data used for the
the median cannot be used as an estimate of the mean. estimation. The uncertainties in the estimators can be repre-
sented by the respective standard deviations, denoted standard
Note also that the two median estimates presented for special errors. If needed, these standard errors can be estimated by
cases in this guidance note are more accurate estimates than the
sample median, because they make use of more information than 1 z2
the sample median does, including the information about distri- se(aˆ 0 )  s   n
bution type. n
 (z
i 1
i  z)2
---e-n-d---of---G-u-i-d-a-n-c-e---n-o-t-e---
s
se(aˆ1 ) 
n
2.4.2 Parameter estimation for dependent soil variables
 (z
i 1
i  z)2
2.4.2.1 Let Y be a soil variable, such as a strength parameter,
whose variation with depth z can reasonably well be modelled 2
se( s 2 )  s 2
as a linear function. Y is then a so-called dependent variable, n2
whereas z is independent. Assume that n observations of pairs
(zi,yi), i = 1,...n, are available from a soil investigation cam- Note that it is the standard error of s2 rather than of s which is
paign. The variation of Y with depth can now be expressed as given here.
Y  a 0  a1 z   2.4.2.5 The estimators in 2.4.2.2 may be recognised as the
results from a least-squares regression analysis.
where the term a0+a1z represents the linear mean variation
with depth and the term  represents the natural variability of 2.4.2.6 Situations exist for which the natural variability of the
Y about the mean, and where z denotes the depth below the soil considered soil parameter Y is such that the standard deviation
surface. The coefficients a0 and a1 represent the surface inter-  of the random variability term  is not a constant as assumed
cept and the depth gradient, respectively, of the mean of Y. The in 2.4.2.1 through 2.4.2.4, but rather has a variation with depth.
variability term  has zero mean and its standard deviation is A standard deviation which is proportional with depth is com-
denoted . The standard deviation  is assumed to be constant, monly encountered. For the case that the standard deviation 
i.e. it is assumed to be independent of depth z. Reference is is proportional with the depth z, = kz, the coefficients a0 and
made to Figure 2-6. a1 in the linear expression for the mean of Y can be estimated

DET NORSKE VERITAS

Recommended Practice DNV-RP-C207, October 2010
Page 16

from the n observed data pairs (z,y) as tion were to be installed in the soil deposit rather than a friction
n n pile: The strength observations would have been concentrated in
n w z w y i i i i
a shallow zone within the soil deposit rather than being approxi-
w z y
mately uniformly distributed over the 34 m depth of the soil
i i i
 i 1
n
i 1
deposit.
i 1
w i It is a prerequisite for application of the data set in this example
aˆ1  n
i 1
that the definition of su associated with the data set is consistent
n
( wi z i ) 2 with the definition of su in the design code which is used to pre-
dict the axial pile capacity.
w z  i 1
2
i i n

w
i 1 ---e-n-d---of---G-u-i-d-a-n-c-e---n-o-t-e---
i
i 1
n n

w y i i
 aˆ1  wi xi Undrained shear strength (kPa)

aˆ 0  i 1
n
i 1 0 20 40 60 80

w
0
i
i 1|
5
in which wi = 1/zi2, where zi denotes the depth of the ith obser-
vation yi of Y. 10

2.4.2.7 The proportionality constant k in the expression for the

natural variability term is estimated by 15
Depth (m)
1 n
kˆ   wi ( yi  (aˆ 0  aˆ1 z i )) 2
n  2 i 1
20

25
2.4.2.8 The estimators â0, â1 and k̂ are encumbered with
uncertainty owing to the limited number of data used for the 30
estimation. The uncertainties in the estimators can be repre-
sented by the respective standard deviations, denoted standard
35
errors. If needed, these standard errors can be estimated by
n
Figure 2-7
w z
2
i i Soil Data Set #1: Data for undrained shear strength vs. depth
se(aˆ 0 )  kˆ  n n
i 1
n

 w  w z  (  wi z i ) 2
2

2.4.3.2 The mean value  of su is estimated by the sample

i i i
i 1 i 1 i 1

n
w
se(aˆ 1 )  kˆ  n n
i 1
 w  w z
i i
i 1 i 1
2
se(kˆ 2 )  kˆ 2
n2
Note that it is the standard error of k̂ 2 rather than of k̂ which
is given here.
2.4.3 Calculation example for independent soil variable
2.4.3.1 Soil Data Set #1 consists of 22 observations of the
undrained shear strength su in a 34 m deep soil deposit, see Fig-
ure 2-7. When the ultimate goal is to design a friction pile for
installation in this soil deposit to a depth of about 34 m, a visual
inspection of the data indicates that it will be a reasonable
model to represent the mean strength as a constant value, inde-
pendent of depth, within this soil deposit.
Guidance note:
The application, i.e. primarily the type of foundation, governs
which soil properties are of interest to be estimated. For an axi-
ally loaded friction pile as considered in this example, the mean
strength over the length of the pile is of interest, forming the basis
for prediction of the axial capacity of the pile.
For a shallow foundation to be designed against horizontal slid-
ing or bearing capacity failure, only the strength in a shallow
zone below the soil surface would be of interest, and a subset of
the 22 strength observations in this example, consisting of the 5
or 6 uppermost observations, would have to be subject to inspec-
tion, analysis and interpretation. As the available soil data often
reflect the type and size of foundation to be installed, if this is
known at the time of the soil investigations, it is likely that the
obtained data set would have been different if a shallow founda-
mean of the data, hence
i
1 n
ˆ  s u   su ,i  60.2 kPa
n 22 i 1
 ( wi z i ) 2
2
i The standard deviation  of su is estimated by the sample
i 1
standard deviation, hence
1 22
ˆ  s   (su ,i  su ) 2  10.6 kPa
22  1 i 1
This gives a variance estimate s2 = 112.3 kPa2. The estimated
coefficient of variation is COV = 10.6/60.2 = 0.176.
2.4.3.3 Provided an ideal, accurate method of measuring su
was applied for observation of the 22 data values, s = 10.6 kPa
will be the final estimate of the standard deviation associated
with the natural variability of the undrained shear strength su.
Usually, this is what should be assumed, regardless of the
accuracy of the method of measurement, since this accuracy is
usually either unknown or is difficult to quantify.
2.4.3.4 Provided the accuracy of the method of measuring su
is known, e.g. from a calibration of the test equipment, the
associated uncertainty can in principle be eliminated from the
estimate s, and the final estimate of the standard deviation
associated with the natural variability of the undrained shear
strength su will come out somewhat smaller than the apparent
standard deviation of 10.6 kPa. Consider as an example that the
standard deviation associated with the accuracy of the meas-
urement method is quantified to meas = 6 kPa. For this case,
the net standard deviation associated with the natural variabil-
ity of su becomes reduced to
s  10.6 2  6 2  8.7 kPa
and the estimated coefficient of variation correspondingly
reduces to COV = 8.7/60.2 = 0.145.

DET NORSKE VERITAS


2.4.4 Calculation example for dependent soil variable
2.4.4.1 Soil Data Set #2 consists of 51 observations of the
undrained shear strength which are available from the depth
range 2-39 m in a clay deposit, see Figure 2-8. The data are
obtained by two different types of tests, viz. direct simple shear
tests and vane tests. The data from the vane tests have been
properly transformed to form a consistent population with the
data from the direct simple shear tests, cf. 2.4.6.
2.4.4.2 By a visual inspection, the data are seen on average to
have a linear variation with depth, E su   a0  a1 z .
Undrained shear strength (kPa)
0 20 40 60 80 100
0 be estimated from the data for Y’s dependency on X. The lin-
5
10
15
Depth (m) 20
25
30
35
40
Figure 2-8
Soil Data Set #2: Data for undrained shear strength varying
linearly with depth
The variability in the undrained shear strength about this mean
is characterised by the standard deviation . For an assumption
of constant standard deviation with depth, the formulas in
2.4.2.2 gives the following estimates of the coefficients a0
and a1
â0 = 2.22 kPa and â1 = 2.35 kPa/m
and the standard deviation of the residuals in su is estimated by
the sample standard deviation in 2.4.2.3
ˆ  s  3.76 kPa
Guidance note:
A negative strength intercept â0 = 2.22 kPa at the soil surface
has been estimated. This is unphysical and serves as a reminder
that the linear model for the strength as a function of depth is
valid only within the depth range of the data, i.e. 2-39 m. Within
this range, the estimated coefficients will result in only positive
mean strength values. The example demonstrates that there may
be limitations with respect to extrapolation of statistical estima-
tion results outside the range of the data used for the estimation.
---e-n-d---of---G-u-i-d-a-n-c-e---n-o-t-e---
2.4.5 Outliers
2.4.5.1 Data observations, which are quite far from the bulk of
data in a data set, are called outliers. Before carrying out a sta-
tistical analysis of a data set, the data set should be inspected
for outliers. As a rule of thumb, any data value which is located
more than two standard deviations away from the mean value
of the data set can be considered an outlier. Graphical presen-
tation is an effective means of detection. Rational, more elab-
orate methods for identification of outliers can be found in
Snedecor and Cochran (1989) and Neter et al. (1990).
DET NORSKE VERITAS
Recommended Practice DNV-RP-C207, October 2010
Page 17
2.4.5.2 It is recommended to remove data points that are iden-
tified as outliers. However, one should always justify removal
of identified outliers by physical evidence, e.g. equipment mal-
function, improperly calibrated equipment, and recording and
transcription errors. Also evident physical differences between
the parent test specimen for a data value identified as an outlier
on the one hand and the parent test specimens for the rest of the
data set on the other, such as a significantly different sand con-
tent, may serve to justify the removal of the identified outlier.
2.4.5.3 Caution should be exercised not to remove data points,
which may appear as outliers, but which do not qualify as out-
liers and which should therefore be kept in the data set. An
example of such a data point is a data point far from the bulk
of data points in an (X,Y) plane and far from the line that can
ear relationship assumed between X and Y might simply not be
valid in the region of the (X,Y) plane where the suspicious data
point is located.
2.4.6 Data from more than one source
2.4.6.1 Caution should be exercised when data from two or
more sources are pooled together to form one data set which is
then subject to statistical analysis. For example, soil strength
data may arise from different types of tests on soil samples.
Data from different types of tests have to be converted to the
same reference scale before any statistical analysis of the
pooled data set will be meaningful.
Guidance note:
Data for undrained shear strength of clay are often available from
UU tests and CIU tests. To convert UU test results to obtain
benchmark CIU test results that can be pooled with results from
actual CIU tests, it is common to multiply the strength data from
the UU tests by a factor. This factor is often set equal to 1.2.
Conversion of other types of data before pooling may require
other transformations than simple multiplication by a conversion
factor.
---e-n-d---of---G-u-i-d-a-n-c-e---n-o-t-e---
2.4.7 Bayesian approach to estimation
2.4.7.1 The Bayesian approach to estimation, which capital-
ises on Bayesian statistics and Bayesian updating, forms an
alternative to estimation by the classical statistical approach as
presented in 2.4.1 through 2.4.4. The Bayesian approach to
estimation allows for inclusion of subjective information and
judgment in estimation, e.g. in terms of assumptions of a prior
distribution of an unknown parameter to be estimated. Accord-
ing to the Bayesian approach this distribution is then subse-
quently updated conditional on available sampled data.
2.4.7.2 The Bayesian approach to estimation is useful for esti-
mation in situations where the classical statistical methods
may not necessarily suffice, e.g. when data from different
sources are to be combined. The Bayesian approach also
allows for inclusion of other information than sample data and
it allows for updating of prior information when new addi-
tional information becomes available.
2.4.7.3 For further details of Bayesian estimation reference is
made to Ang and Tang (1975). For further details of Bayesian
updating reference is made to Classification Notes No. 30.6.
2.5 Estimation of soil parameters with confidence
2.5.1 Parameter estimation with confidence, independent
variables
2.5.1.1 Let X be a soil variable, such as a strength, for which
n observations exist, x1,...xn. When the characteristic value of
X is defined as the mean value of X, and when the characteris-
tic value is required to be estimated with a confidence of 1,
Recommended Practice DNV-RP-C207, October 2010
Page 18

the following expression for the estimate of the characteristic be applied for the characteristic value
value applies
1 3n
s YC  aˆ 0  aˆ1 z  tn  2 ( )  s  
X C  x  t n 1 ( )  n n2  1
n
in which x is the sample mean of X and s is the sample stand- Table 2-3 Coefficient tn2() for YC defined as mean value of Y
ard deviation of X based on the n observations, and tn1() is n2 1= 0.75 1= 0.90 1 = 0.95
given in Table 2-1. This is based on theory for one-sided con- 2 0.82 1.89 2.92
fidence intervals. The sample mean and sample standard devi-
ation are to be calculated according to expressions given in 5 0.73 1.48 2.02
2.4.1.2 and 2.4.1.3. 10 0.70 1.37 1.81
15 0.69 1.34 1.75
Table 2-1 Coefficient tn1() for XC defined as mean value of X
20 0.69 1.33 1.72
n1 1= 0.75 1= 0.90 1= 0.95 30 0.68 1.31 1.70
2 0.82 1.89 2.92 50 0.68 1.30 1.68
5 0.73 1.48 2.02
10 0.70 1.37 1.81 This expression is a linear approximation to the true expres-
sion, which results from theory and which is somewhat nonlin-
15 0.69 1.34 1.75 ear with respect to depth z. The approximate expression is
20 0.69 1.33 1.72 based on the assumption that the n data pairs are sampled
30 0.68 1.31 1.70 approximately uniformly over the depth range considered. It
50 0.68 1.30 1.68 gives slightly conservative results in the central part of this
depth range, and it is not suitable for extrapolation to charac-
2.5.1.2 When the characteristic value of X is defined as the teristic values outside of this depth range. The coefficient esti-
5% quantile in the distribution of X, and when the characteris- mates â0 and â1 and the sample standard deviation s are to be
tic value is required to be estimated with a confidence of 1, calculated according to the expressions given in 2.4.2.2 and
the following expression for the estimate of the characteristic 2.4.2.3, and tn2() is given in Table 2-3. This is based on the-
value applies ory for one-sided confidence intervals.
X C  x  c1 (n)  s
2.5.2.2 When the characteristic value of Y is defined as the
in which x is the sample mean of X and s is the sample stand- 5% quantile in the distribution of Y, and when the characteris-
ard deviation of X based on the n observations, and c1(n) is tic value is required to be estimated with a confidence of 1,
given in Table 2-2. This is based on theory for tolerance the following expression for the estimate of the characteristic
bounds. The sample mean and sample standard deviation are to value applies
be calculated according to expressions given in 2.4.1.2 and YC  aˆ 0  aˆ 1 z  c1 (n)  s
2.4.1.3.
This expression is a linear approximation to the true, some-
Table 2-2 Coefficient c1(n) for XC defined as the 5% quantile what nonlinear expression that results from theory. It is based
of X on the assumption that the n data pairs are sampled approxi-
mately uniformly over the depth range considered. It gives
n 1= 0.75 1 = 0.90 1  = 0.95 slightly conservative results in the central part of this depth
3 3.17 5.33 7.66 range, and it is not suitable for extrapolation to characteristic
5 2.47 3.41 4.21 values outside of this depth range. The coefficient c1(n) is
10 2.11 2.57 2.91 tabulated in Table 2-4 for various confidence levels 1. For
details about the methodology, reference is made to Ronold
15 1.99 2.34 2.57 and Echtermeyer (1996).
20 1.94 2.21 2.40
30 1.88 2.08 2.22 Table 2-4 Coefficient c1(n) for YC
 1.64 1.64 1.64 defined as the 5% quantile of Y
n 1= 0.75 1= 0.90 1  = 0.95
2.5.1.3 Formulas and tables in 2.5.1.1 and 2.5.1.2 are shown 3 5.2 13.8 27.7
for a single variable such as a strength which can be assumed
constant within a soil layer. Other formulas and tables apply 5 2.88 4.41 5.80
for soil properties that vary linearly with depth, see 2.5.2.1 and 10 2.29 2.95 3.44
2.5.2.2. 15 2.14 2.62 2.95
2.5.2 Parameter estimation with confidence, dependent 20 2.06 2.45 2.70
variables 50 1.88 2.10 2.25
100 1.81 1.96 2.06
2.5.2.1 Let Y be a soil variable, such as a strength, whose var-  1.64 1.64 1.64
iation with depth z on average can reasonably well be modelled
as a linear function. Assume that n observations of pairs (zi,yi), 2.5.2.3 The results presented here for soil properties that vary
i = 1,...n, are available from a soil investigation campaign. The linearly with depth are based on an assumption of a constant
mean variation of Y with depth is expressed as a0+a1z, in standard deviation of the considered soil property with depth.
which the coefficients a0 and a1 can be estimated from the However, some soil properties exhibit a standard deviation
observed data pairs as given in 2.4.2.2.
which is proportional with depth. For this case the coefficient
When the characteristic value of Y is defined as the mean value c1(n) will be a function of depth. It will also depend on how
of Y, and when the characteristic value is required to be esti- the data are distributed over the depth range considered. Tabu-
mated with a confidence of 1, the following expression may lated values for c1(n) are not available.

DET NORSKE VERITAS

 Recommended Practice DNV-RP-C207, October 2010
Page 19

2.5.3 Calculation example for independent soil variable
2.5.3.1 Consider Soil Data Set #1 where n = 22 observations
of the undrained shear strength su in a soil deposit are availa-
ble, see Figure 2-7. The mean value is estimated to be
̂ = 60.2 kPa and the standard deviation is estimated to be
s = 10.2 kPa.
2.5.3.2 A geotechnical problem is encountered for which the
characteristic value of the undrained shear strength is to be
taken as the mean value estimated with 90% confidence.
According to the expression in 2.5.1.1 this characteristic value
becomes
s 10.6
su ,C  ˆ  t n1 ( )   60.2  1.33   57.2 kPa
n 22
emphasis on the statistical interpretation and rather use physical
evidence to establish the characteristic value as a physical lower
bound. As an example, in the case of the strength of a clay, such
physical evidence could consist of knowledge about the thixo-
tropic properties of the clay.
---e-n-d---of---G-u-i-d-a-n-c-e---n-o-t-e---
2.5.5 Minimum sample number
2.5.5.1 From the soil samples obtained during the initial phase
of a soil investigation program, it is possible to determine
whether or not the number of samples collected was sufficient
for the accuracy of prediction required by the design engineer.
If not, then an additional phase of soil investigation will be
necessary so that additional samples can be obtained.

2.5.3.3 Another geotechnical problem is encountered for 2.5.5.2 The prediction accuracy required by the design engi-
which the characteristic value of the undrained shear strength neer may be expressed as a range of values within which the
is to be taken as the 5% quantile with 95% confidence. Accord- true but unknown mean value of the population will lie with a
ing to the expression in 2.5.1.2, this characteristic value probability equal to or greater than a specified value. This
becomes specified probability may be recognised as the confidence and
su ,c  ˆ  c1 (n)  s  60.2  2.36 10.6  35.2 kPa the range of values required for accuracy can then be expressed
as a two-sided confidence interval,
2.5.4 Calculation example for dependent soil variable   s  s 
 x  t n 1 ( 2 )  n ; x  t n 1 ( 2 )  n  ,
2.5.4.1 Consider Soil Data Set #2 where n = 51 observations  
of the undrained shear strength su are available from the depth in which x and s are the sample mean and the sample standard
range 2-39 m in a clay deposit, see Figure 2-8. The mean value deviation, respectively, of the soil parameter in question. The
is represented as a linear function of depth z, E[su] = a0+a1z. confidence is 1  and tn1(/2) is given in Table 2-5.
The coefficients in the expression for E[su] are estimated to be
â0 = 2.22 kPa and â1 = 2.35 kPa/m, respectively, and the Table 2-5 Coefficient tn1(/2)
standard deviation of the residuals in su is estimated
n 1 1 = 0.80 1  = 0.90 1 = 0.95
by ̂  s = 3.76kPa.
2 1.89 2.92 4.30
2.5.4.2 Consider the situation that the characteristic value of 3 1.64 2.35 3.18
su is to be taken as the mean value of su estimated with 95% 4 1.53 2.13 2.78
confidence. According to the expression in 2.5.2.1, the charac- 5 1.48 2.02 2.57
teristic su profile then becomes
8 1.40 1.86 2.31
1 3n 10 1.37 1.81 2.23
s u , C  aˆ 0  aˆ 1 z  t n  2 ( )  s   2
n n 1 15 1.34 1.75 2.13
1 3  51 20 1.33 1.72 2.09
 2.22  2.35 z  1.68  3.76   30 1.31 1.70 2.04
51 512  1
50 1.30 1.68 1.96
 3.99  2.35 z
Guidance note:
2.5.4.3 Consider the situation that the characteristic value of Assume as an example that four observations of the undrained
su is to be taken as the 5% quantile of su estimated with 95% shear strength are available from tests on four soil samples from
confidence. According to the expression in 2.5.2.2, the charac- a stiff clay deposit: 93, 100, 104 and 107 kPa. The minimum
teristic su profile then becomes number of soil samples required so that the true, unknown mean
s u ,C  aˆ 0  aˆ1 z  c1 (n)  s value will be within 5% of the mean test result with at least 90%
probability is sought.
 2.22  2.35 z  2.25  3.76
The mean value estimate is x = 101 kPa and the standard devia-
 10.68  2.35 z tion estimate is s = 6.06 kPa. The limits of the 90% confidence
Guidance note: interval become 101 2.35 6.06/4 = 93.9 to 108.1 kPa, i.e.
The resulting expression for su,C gives negative characteristic some 7% on either side of the mean.
shear strength in a depth range between the soil surface and about Four samples are thus not sufficient if the true mean shall be
4.5 m depth. This is unphysical and may be interpreted as an indi- within 5% of the mean test result. This is an interval from 96 to
cation that the normal distribution assumption for the undrained 106 kPa. By combining the expression for the confidence inter-
shear strength may not necessarily be a good model in the upper- val with Table 2-5, it can be determined that an interval equal to
most part of the soil deposit. In the actual case, if it is maintained or narrower than the required interval from 96 to 106 kPa will be
that the 5% quantile shall be estimated with 95% confidence, the achieved when the number of samples is increased by two to a
distribution assumption should be revisited and a truncated nor- total of six samples, i.e. two more samples must be obtained and
mal distribution should be considered instead of the normal dis- tested.
tribution before the characteristic strength is estimated with the
specified confidence. Otherwise, it may be considered to put less ---e-n-d---of---G-u-i-d-a-n-c-e---n-o-t-e---

DET NORSKE VERITAS

Recommended Practice DNV-RP-C207, October 2010
Page 20
2.5.6 Bayesian approach to estimation with confidence
2.5.6.1 As an alternative to the classical statistical approach to
estimation with confidence presented in 2.5.1 through 2.5.4, a
Bayesian approach to estimation with confidence may be
applied. This requires that a suitable loss function be chosen.
The choice of loss function governs the confidence associated
with the estimate that results from the Bayesian estimation.
The choice of the right loss function is thus crucial in order to
arrive at an estimate that has the desired confidence.
For details of Bayesian estimation with confidence, reference
is made to Ang and Tang (1984).
2.6 Combination of variances
2.6.1 Combination rules
2.6.1.1 Variances associated with different uncertainty
sources are combined by simple addition when the uncertainty
sources are additive and statistically independent.
2.6.1.2 Let X be a soil variable such as a strength. X will
exhibit a natural variability from point to point within a soil
deposit. The mean and variance of X are estimated on the basis
of n observations by measurements on soil samples, so the esti-
mates of the mean and variance of X are encumbered with
measurement uncertainty as well as with statistical uncer-
tainty. The measurement uncertainty and the statistical uncer-
tainty are both epistemic uncertainties which are independent
of the natural variability of X. In addition the measurement
uncertainty and the statistical uncertainty can be assumed to be
statistically independent.
2.6.1.3 The observed variance obs which is estimated from 2,
the n observations of X, forms an apparent variance which is a
combination of a variance X2 associated with natural variabil-
ity and a variance meas2 associated with measurement uncer-
tainty,
 obs   X   meas
2 2
The total variance associated with the determination of X will
include a variance stat2 associated with statistical uncertainty
owing to limited data, hence
 tot   obs   stat   X   meas   stat
2 2 2 2
If some model is applied to transform n measurements on soil
samples to n observations of X before the estimates of the
mean and variance of X are established from the data, a model
uncertainty will be present. Examples of this are vane test data
which are transformed to undrained shear strengths and CPT
readings which are converted to undrained shear strengths. For
such cases, the variance model2 associated with the applied
conversion model are included in the observed variance obs2
and therefore appears in the expressions in the same manner as
meas2, hence
 obs   X   meas   model
2 2 2 2
and
 tot   obs   stat   X   meas   model   stat
2 2 2 2 2 2
2.6.1.4 When the variance meas2 associated with the meas-
urement uncertainty can be quantified, it can in principle be
eliminated from the observed variance obs2 and the variance
X2 associated with the natural variability of X can be isolated,
 X   obs   meas
2 2
In the case that also a model uncertainty is involved in estab-
lishing the n observations of X, the variances associated with
both the measurement uncertainty and the model uncertainty
need to be quantified in order that the variance associated with
the natural variability of X can be isolated,
 X   obs   meas   model
2 2 2 2
Guidance note:
It can be dangerous to perform the reduction in the observed var-
iance for possible measurement and model uncertainties.
Caution should be exercised not to carry out the favourable
reduction of the observed variance obs2 by variances associated
with measurement and model uncertainty, unless it is absolutely
certain that these uncertainties are present in the data and their
variances can be properly quantified and trusted. Otherwise, one
may be unfortunate to underestimate X2 from obs2.
---e-n-d---of---G-u-i-d-a-n-c-e---n-o-t-e---
2.6.1.5 The autocorrelation function (r) for the variable X
considered as a random field over the soil deposit can in prin-
ciple be estimated from spatial observations of X. The estimate
of this autocorrelation function can be applied to provide infor-
mation which can be used to isolate the variance X2 from the
observed variance obs2.
The autocorrelation function shall theoretically attain the value
= 1 for lag r = 0. However, when the autocorrelation function
is estimated from measured values of X for various lags r > 0,
the results can be extrapolated to estimate the value of  for
r = 0. The result of this exercise will usually be that b for
r  0, where b < 1. The value of the limit b being less than 1
reflects the presence of measurement uncertainty and a possi-
ble model uncertainty as described in 2.6.1.2 and 2.6.1.3, and
the variance associated with the natural variability of X can be
interpreted as
 X  b   obs
2 2
Caution must be exercised to avoid underestimation of b and
thereby of X2 when b is extrapolated from available estimated
(r) values. Estimated (r) values will often be encumbered
with large uncertainties, which together with the nonlinearity
of (r) with r makes the extrapolation difficult.
2.6.2 Calculation example for independent soil variable
2
2.6.2.1 Consider Soil Data Set #1 where n = 22 observations
of the undrained shear strength su in a soil layer are available,
see Figure 2-7. The mean value of su is estimated to be
̂ = 60.2 kPa and the standard deviation of su is estimated to
2 2 be s = 10.2 kPa.
2.6.2.2 There is statistical uncertainty associated with the esti-
mates of the mean value and the standard deviation of su
because only 22 observations of su were available for the esti-
mation. The standard error in the estimate of the mean value is
10.6
se( s u )   2.3 kPa
22
and the standard error in the estimate of the standard deviation
is
2.22  1
se( s )  10.6  1.25 kPa
4  22
which is based on a kurtosis estimate of 2.22.
2
2.6.2.3 When the soil variable su is not known in full by its
distribution with known mean value and known standard devi-
ation, but only to the extent that the mean value and standard
deviation have been estimated from limited data, it has an asso-
ciated uncertainty which stems not only from its natural varia-
bility but also from the statistical uncertainty associated with
2
the estimates of the mean and the standard deviation. The esti-
mate of the total variance associated with su can then be
expressed as
ˆ tot  s 2  se( s u ) 2  se( s ) 2
2
 10.6 2  2.3 2  1.25 2
 10.9 2 kPa 2
DET NORSKE VERITAS
 Recommended Practice DNV-RP-C207, October 2010
Page 21

2.7 Geostatistics Guidance note:

When Su is constant with depth and when the vertical correla-
2.7.1 Spatial averaging tion structure of the strength field can be represented by the quad-
ratic exponential decay model, the variance reduction factor can
2.7.1.1 When the application of the information about the soil be expressed as
variable X implies that a spatial average of X over large soil 2    L  1  1 2   L2   
volumes or areas shall be used rather than X itself, then the full   LR   2     R exp  2   1 
variance X2 shall not be used in calculations but rather a L2    R  2  2   R   
reduced variance in which L is the length of the pile, R is the vertical correlation
X
2
  X
2
length of the undrained shear strength su, and  is the standard
spatial
normal cumulative distribution function. See Figure 2-9.
in which < 1 is a variance reduction factor which reflects the ---e-n-d---of---G-u-i-d-a-n-c-e---n-o-t-e---
extent of the spatial averaging and which depends on the auto-
correlation function for X and the geometry of the involved
soil volume or area. In applications, this reduced variance
needs to be superimposed by the variance stat2 associated
with possible statistical uncertainty owing to limited data and
by a variance applmodel2 associated with a possible model
uncertainty involved with the application. When full averaging
takes place, i.e. when local fluctuations of X from point to
point average out completely over the involved volume or
area, then  = 0, and the only variance which is left is
stat2 + applmodel2. An example of spatial averaging over a
continuous soil volume is given in 2.7.2.
Solutions for the variance reduction factor  for a number of
geometries can be found in Ronold (1990).
2.7.1.2 Spatial averaging may also be relevant for problems
and mechanisms which involve averages or sums over a Figure 2-9
number of discrete elements in space rather than averages or Variance reduction factor vs. pile length L
integrals over continuous soil volumes. and vertical correlation length R

Guidance note:
An example of a problem which involves spatial averaging of
soil properties over a number of discrete elements in space is the
lateral capacity of the piled foundation of an offshore jacket plat-
form. This capacity comes about as N times the spatial average
of the lateral capacities of the N jacket piles that constitute the
foundation. The spatial average of the lateral pile capacities will
2.7.2.3 Statistical uncertainty associated with the estimation
of the mean Su and the standard deviation Su of su comes in
addition when the total uncertainty associated with the predic-
tion of Qaxial is sought after. Through the expression for Qaxial,
the mean Su of su appears to govern the mean value of Qaxial,
L

E Qaxial   D    su ( z )dz

depend on the autocorrelation function for the soil strength that
produces the lateral pile capacities, it will depend on the geomet-
rical configuration of the N pile positions and it will depend on z 0

the separations between these positions in the horizontal plane.

---e-n-d---of---G-u-i-d-a-n-c-e---n-o-t-e---
2.7.2 Example of spatial averaging
2.7.2.1 An example of an application, where the spatial aver-
age of a soil variable is involved rather than the variable itself,
is the axial capacity of a friction pile in clay. Several models
exist for prediction of this capacity, including the  method.
By this model, the spatial average of the undrained shear
strength su over the length L of the pile is involved as part of
the expression for the axial capacity Qaxial, viz.
L Su and Su can be estimated from the soil data obtained
Qaxial  D   s u ( z )dz
z 0
 L L
 3) The position of the pile relative to soil borings and CPTs
 D     s ( z )dz    ( z )dz 
 z 0 
u
z 0
in which  is a dimensionless factor and D is the pile diameter.
Su denotes the mean value of su and  is a zero-mean term
which represents the variability of su about Su.
2.7.2.2 The reduced variance associated with the spatially
averaged undrained shear strength su used for prediction of
Qaxial is Su,spatial2 = Su2 and comes about as the variance
of the spatial average term
1 L
L z 0
 ( z )dz
2.7.2.4 The estimation of the mean Su and the standard devi-
ation Su of su and the determination of the involved estima-
tion uncertainties depend on where the friction pile is located
relative to available soil borings and CPTs. Three different sit-
uations are considered:
1) The position of the pile relative to soil borings and CPTs
is unknown, and Su and Su can be estimated from all soil
data available from the soil deposit in which the pile is to
be installed.
2) The position of the pile relative to soil borings and CPTs
is known and coincides with one of these borings or CPTs.
from the boring or the CPT which is located in the pile
position.
is known, but does not coincide with any of these borings
or CPTs. Su and Su can be estimated by spatial interpo-
lation between soil data from adjacent borings and CPTs,
e.g. by means of a kriging technique. See 2.7.3.
A variance associated with the variability in predictions by the
model for Qaxial from case to case comes in addition to the
reduced variance Su,spatial2 and the statistical variance stat2.
2.7.3 Spatial estimation – kriging
2.7.3.1 Let X denote a soil variable, such as a strength, which
forms a random field over a soil deposit. X has been measured
in a number of positions within the soil volume, but is in prin-

DET NORSKE VERITAS

Recommended Practice DNV-RP-C207, October 2010
Page 22
ciple unknown in all other positions. It is assumed that the
autocorrelation function for X can be established based on the
available observations of X.
2.7.3.2 Kriging is an estimation technique which, based on the
measured values of X in conjunction with the autocorrelation
function, allows for estimation of the value of X in positions
where X has not been measured. For each specified position of
estimation, the kriging technique provides both an estimate of
X and a standard error in the estimate, accounting for spatial
correlations.
2.7.3.3 Assume that X is observed in n positions, si, i = 1,...n,
within the soil volume V and let xi denote the observed value
of X in the position si. Assume also that the autocorrelation
function (r) for X is available such that the covariance func-
tion C(r) = 2(r) can be established. Here, r denotes the lag
between two positions s and s’, r = |ss’|. The point value of X
in the unsampled position s is to be estimated.
2.7.3.4 The simple kriging estimate of the single point realisa-
tion X(s) in the position s is
n 2.7.4 Calculation example of kriging
xˆ (s)   i xi
i 1
where i, i = 1,...n, are weighting factors, which can be deter-
mined as the solution of the n equations
n
  C (| s
j 1
j i  s j |)  C (| s  s i |) ; i  1,...n
2.7.3.5 The ordinary kriging estimate of the single point real-
isation x(s) of X in the position s is
n Guidance note:
xˆ (s)   i xi
i 1
where i, i = 1,...n, are weighting factors, which can be deter-
mined as the solution of the n equations
n
  C (| s
j i  s j |)  k  C (| s  s i |) ; i  1,...n
j 1
with the unbiasedness constraint
n 2.7.4.2 The measurement positions and the strength measure-

i 1
i 1
2.7.3.6 The point estimate xˆ (s ) is encumbered with uncer-
tainty because it is based on a limited number of observations
of X in a limited number of positions located some distance
away from the point of estimation. Regardless of whether the
estimate xˆ (s) is established by simple kriging or by ordinary
kriging, the standard error of the estimate is
n n
se( xˆ (s))  C (0)     i  j C (| s i  s j |)  2  i C (| s  s i |)
i 1 j 1
n n
   (0)     i  j  (| s i  s j |) 2  i  (| s  s i |)
i 1 j 1
2.7.3.7 The expressions for the kriging estimates and their
standard error are based on assumptions of isotropic autocor-
relation and covariance functions. When these functions are
anisotropic, e.g. when 3-D problems are encountered and dif-
ferent correlation lengths apply vertically and horizontally, the
expressions will become modified as the arguments of these
functions, which here appear as a scalar measure of lag,
become a vector of lag in space.
2.7.3.8 Kriging can be generalised to apply to prediction of
the spatial average of the field X over the soil volume V rather
than just estimating the point value in a particular position s.
Kriging, which is here presented for the case of a stationary
field X with constant mean and variance, can also be general-
ised to kriging with a trend for the case that the field X has a
mean that has a variation with location.
2.7.3.9 Kriging can be applied to planning of soil investiga-
tions, e.g. determination of the most optimal position for an
additional soil boring when an extra soil boring is to be carried
out to supplement an existing array of soil borings in a soil
deposit. In this context the most optimal position is the position
that will provide as much new knowledge and as little redun-
dant knowledge about the random field X as possible.
Guidance note:
When an additional soil boring is to be carried out, all relevant
information, including knowledge of the local geology and
results from geophysical surveys, will have to be brought into the
picture in addition to the results from the kriging analysis, before
the final position of the additional boring can be decided.
---e-n-d---of---G-u-i-d-a-n-c-e---n-o-t-e---
2.7.3.10 Kriging can also be applied to mapping of soil layer
boundaries.
2.7.4.1 The undrained shear strength su in a soil layer is meas-
ured in 3 positions which form 3 of the corners of a horizontal
square whose side length is r = 15 m. The undrained shear
strength in the fourth corner of the square is unknown but can
be estimated by means of kriging on the basis of the 3 observed
values. The autocorrelation function for the strength field is
assumed to be (r) = exp((r/R)2) with R = 30 m. Possible
measurement uncertainties and uncertainties associated with
sample disturbance are disregarded.
Kriging can be applied regardless of the spatial geometry of the
data and does thus not presume a particular configuration of the
data points such as the four corners of a square used in this exam-
ple. What matters is the distances between the data points them-
selves and between the data points and the point of prediction,
not the absolute positions of these points.
---e-n-d---of---G-u-i-d-a-n-c-e---n-o-t-e---
ments are
Position no. Position coordinates s = (x,y) su (kPa)
x (m) y (m)
1 0 15 80
2 0 0 85
3 15 0 75
n 2.7.4.3 Three values of the correlation coefficient  are
required in the kriging equations, viz. (0) = 1.0,
i 1
(15) = 0.779, and (21.2) = 0.607.
n
2.7.4.4 Solution of the set of 3 kriging equations leads to the
i 1 weights 1 = 3 = 0.779 and 2 = 0.607, and the simple krig-
ing estimate of su in the unsampled position (x,y) = (15,15)
becomes
3
sˆ u (15,15)    i s u , i  69.2 kPa
i 1
The ratio of the estimation variance se( sˆu ) 2 to the variance
Var[su] of the undrained shear strength field su is
se( sˆu (15,15)) 2 3 3
 1   i  j  (| s i , s j |)
Var s u  i 1 j 1
3
 2 i  (| s i , (15,15) |)
i 1
 0.155
DET NORSKE VERITAS

2.7.4.5 It appears that the simple kriging applied to point esti-
mation of the su field in the position s = (15,15) implies an
updating of the distribution of su in this position from a prior
distribution to a posterior distribution by determination of the
conditional distribution from observations of su in neighbour-
ing locations. In particular it appears that for the present exam-
ple the updating implies a reduction in the variance of the point
estimate of su by 84%.
The prior information consist of the mean estimate ̂ = 80 kPa
and the standard deviation estimate ˆ = 5 kPa, both based on
the three available observations of su given in 2.7.4.2. These
are best estimates for the undrained shear strength and the
standard deviation of the undrained shear strength, respec-
tively, in any point within the undrained shear strength field
regardless of location.
When the additional information about geometry and autocor-
relation is included and accounted for through the kriging, the
estimate of the undrained shear strength in the considered par-
ticular point of prediction becomes updated to 69.2 kPa as out-
lined in 2.7.4.3 and 2.7.4.4, and the standard deviation of this
estimate becomes
se( sˆu (15,15))  0.155  5  2.0 kPa
which is a significant reduction compared to the prior estimate
of 5 kPa.
2.7.4.6 It is of interest to compare the simple kriging result
presented in 2.7.4.4 with the result of estimation by ordinary
kriging. Ordinary kriging implies that an unbiasedness con-
straint comes into force and solution of the 3 kriging equations
of the example leads to altered weights 1= 3 = 0.875 and
2 = 0.751. The ordinary kriging estimate of su in the unsam-
pled position (x,y) = (15,15) hence becomes
3 ing variables X.
sˆ u ,OK (15,15)    i s u ,i  71.8 kPa
i 1
The difference between the simple kriging estimate 69.2 kPa
and the ordinary kriging estimate 71.8 kPa illustrates that sim-
ple kriging can be biased. In the present case the bias could
well be associated with the fact that the estimation is based on
as little as 3 observed field values and that it is an extrapolation
rather than an interpolation. The difference between the two
estimates demonstrates the added value provided by the unbi-
asedness constraint of the ordinary kriging.
2.8 Full probability distribution representation of
soil properties with examples of application
2.8.1 General
2.8.1.1 A number of applications where uncertainties in soil
properties are involved require full probability distribution
representation of the soil properties. Such applications include
structural reliability analyses, by which the probability of fail-
ure for a foundation can be calculated when all governing load
and soil variables and their probability distributions are mod-
elled. Full probability distribution representation of soil prop-
erties usually consists of a parametric distribution model, i.e. a
generic distribution type, such as the normal distribution, and
as many distribution parameters as required, e.g. mean value
and standard deviation.
2.8.1.2 The most commonly encountered generic distribution
types for soil properties are given in 2.3.3 and 2.3.4.
2.8.2 Application to geotechnical reliability analyses
2.8.2.1 In geotechnical design, the reliability of a foundation
can be evaluated with respect to relevant failure modes. For
DET NORSKE VERITAS
Recommended Practice DNV-RP-C207, October 2010
Page 23
each such failure mode, the foundation and its engineering
behaviour can be described by a set of stochastic basic varia-
bles grouped into one vector X, including, e.g., the soil
strength, the soil stiffness, the foundation geometry, and the
loading. All of these variables are stochastic in the sense that,
owing to natural variability and possible other uncertainties,
they may take on random realizations according to some prob-
ability distribution.
2.8.2.2 For the considered failure mode, the possible realisa-
tions of X can be separated in two sets; namely the set for
which the foundation will be safe, and the set for which it will
fail. The surface between the safe set and the failure set in the
space of basic variables is denoted the limit state surface, and
the reliability problem is conveniently described by a so-called
limit state function g(X) which is defined such that
  0 for X in safe set

g ( X)  0 for X on limit state surface
 0 for X in failure set
The limit state function is usually based on some mathematical
engineering model for the considered limit state, based on the
underlying physics, and expressed in terms of the governing
load and resistance variables.
2.8.2.3 The failure probability is the probability content in the
failure set

PF  P g ( X)  0    f X ( x ) dx
g ( X ) 0
where fX(x) is the joint probability density function for X and
represents the uncertainty and natural variability in the govern-
2.8.2.4 The complementary probability PS = 1PF is referred
to as the reliability and is sometimes also denoted the probabil-
ity of survival.
2.8.2.5 The reliability may be expressed in terms of the relia-
bility index,
= 1(PF),
where  is the standardized cumulative normal distribution
function.
2.8.2.6 The failure probability, the reliability, and the reliabil-
ity index are all suitable measures of the foundation safety with
respect to the considered failure mode.
2.8.2.7 For the simple example that X consists of two varia-
bles, the load L and the resistance R, and the limit state function
can be specified as g(X) = RL, the failure probability becomes
a simple convolution integral
PF  PR  L  0
 f
R  L0
R (r ) f L (l )drdl
 l
 f
 
R (r ) f L (l )drdl

  FR (l ) f L (l )dl

where fR and fL are the probability density functions of R and
L, respectively, and fR(r) = dFR(r)/dr, where FR is the cumula-
tive distribution function of R. Reference is made to Figure 2-
10.
Recommended Practice DNV-RP-C207, October 2010
Page 24
Figure 2-10
Probability density functions and characteristic values for load L
and resistance R
2.8.2.8 The reliability index  can be solved in a structural
reliability analysis by means of a reliability method which can
be any amongst several available methods, including numeri-
cal integration, analytical first- and second-order reliability
methods (FORM/SORM), and simulation methods. Simula-
tion methods include the well-known Monte Carlo simulation.
Reference is made to Madsen et al. (1986). Some of these
methods are approximate methods, which will lead to approx-
imate results for the reliability index. Numerical integration is
usually only feasible when X consists of very few stochastic
variables such as in the example in 2.8.2.7. Analytical first-
and second-order solutions to the failure probability are often
sufficiently accurate, and they are advantageous to simulation
results when the failure probabilities are small.
Guidance note:
When the reliability index as defined in 2.8.2.5 is estimated by a
first-order reliability method (FORM), the estimate that results
from the estimation is a first-order second-moment reliability index
according to Hasofer-Lind’s definition, see Madsen et al. (1986).
It is important to distinguish the reliability index as defined in
2.8.2.5 from the so-called simple reliability index. The simple
reliability index is defined as the ratio between the mean value
and the standard deviation of the limit state function. The simple
reliability index is not invariant with the choice of limit state
function, it will in general be different from the reliability index
as defined in 2.8.2.5, and it is not recommended for use.
---e-n-d---of---G-u-i-d-a-n-c-e---n-o-t-e---
2.8.2.9 A useful byproduct of a structural reliability analysis
by these methods is the so-called design point x*. This is a
point on the limit state surface and is the most likely realisation
of the stochastic variables X at failure. Another useful byprod-
uct is the uncertainty importance factors for X. The uncertainty
importance factor for the ith stochastic variable Xi in X
roughly gives the fraction of the total uncertainty which is
caused by the uncertainty in Xi.
2.8.2.10 Once a target failure probability is specified and once
characteristic values of governing load and resistance variables
for a foundation are defined, e.g. as quantiles in the respective
probability distributions of these variables, structural reliabil-
ity analyses can be applied to calibrate the necessary require-
ments to the load and material factors that are part of the design
rule for the foundation.
2.9 Design soil parameters
2.9.1 General
2.9.1.1 Soil parameters for use in design are usually given in
one or more of the following ways: Characteristic value of the
soil property in question, best estimate of the soil property, and
upper and lower bounds of the soil property. Soil parameters
given in these ways for use in design should be established
according to the methodologies presented in 2.4.1 through
2.4.4 and 2.5.1 through 2.5.4.
DET NORSKE VERITAS
2.9.1.2 So-called design profiles of characteristic values for
soil properties are often established with reference to a specific
application, i.e. with reference to a specific type and size of
foundation, and will not necessarily be valid for other applica-
tions.
Guidance note:
The reason why design profiles established for specific applica-
tions will not necessarily be valid for other applications is that the
definition of characteristic value may be different for different
applications. This depends on the degree to which averaging of
soil properties can be assumed to take place. For long friction
piles, the degree of averaging may be large, whereas the degree
of averaging may be a lot smaller for end-bearing piles and
smaller shallow foundations.
The need for close contact between the soil investigation contrac-
tor, who shall deliver a soil investigation report and who is most
familiar with the assumptions made in establishing the character-
istic values and with the uncertainties present, and the geotechni-
cal designer who shall use this report for foundation design, is
evident, but is in practice – unfortunately – often neglected.
Caution should be exercised when using prescribed design pro-
files for soil properties. Some standards refer to both characteris-
tic profiles and design profiles and distinguish between them by
means of the material factor. Other standards refer only to one of
the profiles, and for some standards, which refer to the design
profile only, the design profile is not necessarily factored, i.e. the
design profile may consist of unfactored characteristic values
rather than characteristic values divided by the material factor.
Hence, prior to using a prescribed design profile, one should
always make sure to check whether the soil property values asso-
ciated with the profile have been divided by a material factor or
not.
---e-n-d---of---G-u-i-d-a-n-c-e---n-o-t-e---
2.9.2 Characteristic value
2.9.2.1 Characteristic values are used to represent soil proper-
ties such as soil shear strength. The definition of a characteris-
tic value is much governed by the design problem in question
and by the geometry. A definition as the mean value of the
property typically applies in cases where local fluctuations of
the soil property can be assumed to average out over large soil
volumes, such as in the case of the axial capacity of long fric-
tion piles. A definition as a lower-tail quantile in the distribu-
tion of the property typically applies in cases where a local soil
strength is governing, such as in the case of the tip resistance
of an end-bearing pile.
Guidance note:
In foundation design, the characteristic value of a soil property is
always used together with a partial safety factor. The character-
istic value and the partial safety factor form a pair. This implies
that a partial safety factor, which is meant for use together with a
characteristic value defined as the mean value of a soil property,
cannot be used together with a characteristic value defined as a
low percentile of that soil property.
---e-n-d---of---G-u-i-d-a-n-c-e---n-o-t-e---
2.9.3 Best estimate
2.9.3.1 Best estimates of soil properties are central unbiased
estimates with lowest possible standard errors. Best estimates
are normally used for assessment of serviceability limit states,
i.e. whenever problems are encountered for which predictions
of the expected foundation behaviour are of interest. An exam-
ple of this is the prediction of the expected consolidation set-
tlements for which best estimate values of governing soil
deformation parameters are required, e.g. inferred from oed-
ometer tests. This applies to prediction of absolute consolida-
tion settlements as well as to prediction of differential
settlements. For prediction of best estimates of differential set-
tlements, the horizontal correlation lengths of the governing
soil deformation parameters are of importance.

2.9.4 Lower and upper bounds
2.9.4.1 Soil reports often specify lower and upper bound val-
ues for use as characteristic values in design. Lower bounds are
usually meant for design against the ultimate limit state where
low strengths are unfavourable. Upper bounds are usually
meant for considerations where large strengths are unfavoura-
ble such as evaluations of skirt penetration resistance for
installation of skirted foundations and pile driving resistance
for installation ability of piles by means of particular pile driv-
ing hammers. Upper bounds are also of relevance for evalua-
tion of pile driving resistance for assessment of fatigue in the
pile material. The designer should always make sure that pre-
sented lower and upper bounds for soil properties meet the
requirements to characteristic values set forth in the codes and
standards which are used in design.
For design of structures subjected to cyclic loading or influ-
enced by dynamic behaviour, it may be necessary to perform
sensitivity studies for both lower bound values and upper
bound values for relevant soil properties for the supporting
foundation soils.
2.10 References and literature
2.10.1 List of references
2.10.1.1 Ang. A., and W.H. Tang, Probability Concepts in
Engineering Planning and Design, Vol. I, Basic Principles,
John Wiley and Sons, New York, N.Y., 1975.
2.10.1.2 Ang. A., and W.H. Tang, Probability Concepts in
Engineering Planning and Design, Vol. II, Decision, Risk, and
Reliability, John Wiley and Sons, New York, N.Y., 1984.
2.10.1.3 Madsen, H.O., N.C. Lind, and S. Krenk, Methods of
DET NORSKE VERITAS
Recommended Practice DNV-RP-C207, October 2010
Page 25
Structural Safety, Prentice Hall Inc., Englewood Cliffs, N.J.,
1986.
2.10.1.4 Neter, J., W. Wasserman, and M.H. Kutner, Applied
Linear Statistical Models, Richard D. Irwin, Inc., Homewood,
Ill., 1990.
2.10.1.5 Ronold, K.O., “Random Field Modeling of Founda-
tion Failure Modes,” Journal of Geotechnical Engineering,
ASCE, Vol. 116, No. 4, pp. 554-570, 1990.
2.10.1.6 Ronold, K.O., and A. Echtermeyer, “Estimation of
fatigue curves for design of composite laminates,” Compos-
ites, Part A: Applied Science and Manufacturing, Elsevier,
Vol. 27A, No. 6, pp. 485-491, 1996.
2.10.1.7 Snedecor, G.W., and W.G. Cochran, Statistical
Methods, Iowa State University Press, Ames, Iowa, 1989.
2.10.2 Other relevant literature
2.10.2.1 Benjamin, J.R., and C.A. Cornell, Probability, Statis-
tics, and Decision for Civil Engineers, McGraw-Hill, Inc.,
New York, N.Y., 1970.
2.10.2.2 Cressie, N., Statistics for Spatial Data, John Wiley
and Sons, New York, N.Y., 1991.
2.10.2.3 Isaaks, E.H., and R.M. Shrivastava, Applied Geosta-
tistics, Oxford University Press, New York, N.Y., 1989.
2.10.2.4 Journel, A.G., and C.J. Huijbregts, Mining Geostatis-
tics, Academic Press, London, England, 1978.
2.10.2.5 Vanmarcke, E., Random Fields, Analysis and Synthe-
sis, The MIT Press, Cambridge, Mass., 1983.
Recommended Practice DNV-RP-C207, October 2010
Page 26

APPENDIX A
TABLES OF PROBABILITY DISTRIBUTIONS

A.1 Table of the Student’s t distribution

Table A-1 gives the 1 quantiles t in the Student’s t distri-
bution with n degrees of freedom, i.e. P[T < t] = 1

Table A-1 Quantiles in the Student’s t distribution

1 
n
0.75 0.90 0.95 0.98 0.99
1 1.000 3.078 6.314 15.895 31.821
2 0.816 1.886 2.920 4.849 6.965
3 0.765 1.638 2.353 3.482 4.541
4 0.741 1.533 2.132 2.999 3.747
5 0.727 1.476 2.015 2.757 3.365
6 0.718 1.440 1.943 2.612 3.143
7 0.711 1.415 1.895 2.517 2.998
8 0.706 1.397 1.860 2.449 2.896
9 0.703 1.383 1.833 2.398 2.821
10 0.700 1.372 1.812 2.359 2.764
11 0.697 1.363 1.796 2.328 2.718
12 0.695 1.356 1.782 2.303 2.681
13 0.694 1.350 1.771 2.282 2.650
14 0.692 1.345 1.761 2.264 2.624
15 0.691 1.341 1.753 2.249 2.602
16 0.690 1.337 1.746 2.235 2.583
17 0.689 1.333 1.740 2.224 2.567
18 0.688 1.330 1.734 2.214 2.552
19 0.688 1.328 1.729 2.205 2.539
20 0.687 1.325 1.725 2.197 2.528
21 0.686 1.323 1.721 2.189 2.518
22 0.686 1.321 1.717 2.183 2.508
23 0.685 1.319 1.714 2.177 2.500
24 0.685 1.318 1.711 2.172 2.492
25 0.684 1.316 1.708 2.167 2.485
26 0.684 1.315 1.706 2.162 2.479
27 0.684 1.314 1.703 2.158 2.473
28 0.683 1.313 1.701 2.154 2.467
29 0.683 1.311 1.699 2.150 2.462
30 0.683 1.310 1.697 2.147 2.457
35 0.682 1.306 1.690 2.133 2.438
40 0.681 1.303 1.684 2.123 2.423
45 0.680 1.301 1.679 2.115 2.412
50 0.679 1.299 1.676 2.109 2.403
60 0.679 1.296 1.671 2.099 2.390
70 0.678 1.294 1.667 2.093 2.381
80 0.678 1.292 1.664 2.088 2.374
100 0.677 1.290 1.660 2.081 2.364
 0.674 1.282 1.645 2.054 2.326

DET NORSKE VERITAS

 Recommended Practice DNV-RP-C207, October 2010
Page 27

A.2 Table of the normal distribution

Table A-2 gives the value of the cumulative normal distribu-
tion function (x) for given quantile x.

Table A-2 The cumulative normal distribution function

x 0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
0.0 0.5000 0.5040 0.5080 0.5120 0.5160 0.5199 0.5239 0.5279 0.5319 0.5359
0.1 0.5398 0.5438 0.5478 0.5517 0.5557 0.5596 0.5636 0.5675 0.5714 0.5753
0.2 0.5793 0.5832 0.5871 0.5910 0.5948 0.5987 0.6026 0.6064 0.6103 0.6141
0.3 0.6179 0.6217 0.6255 0.6293 0.6331 0.6368 0.6406 0.6443 0.6480 0.6517
0.4 0.6554 0.6591 0.6628 0.6664 0.6700 0.6736 0.6772 0.6808 0.6844 0.6879
0.5 0.6915 0.6950 0.6985 0.7019 0.7054 0.7088 0.7123 0.7157 0.7190 0.7224
0.6 0.7257 0.7291 0.7324 0.7357 0.7389 0.7422 0.7454 0.7486 0.7517 0.7549
0.7 0.7580 0.7611 0.7642 0.7673 0.7704 0.7734 0.7764 0.7794 0.7823 0.7852
0.8 0.7881 0.7910 0.7939 0.7967 0.7995 0.8023 0.8051 0.8078 0.8106 0.8133
0.9 0.8159 0.8186 0.8212 0.8238 0.8264 0.8289 0.8315 0.8340 0.8365 0.8389
1.0 0.8413 0.8438 0.8461 0.8485 0.8508 0.8531 0.8554 0.8577 0.8599 0.8621
1.1 0.8643 0.8665 0.8686 0.8708 0.8729 0.8749 0.8770 0.8790 0.8810 0.8830
1.2 0.8849 0.8869 0.8888 0.8907 0.8925 0.8944 0.8962 0.8980 0.8997 0.9015
1.3 0.9032 0.9049 0.9066 0.9082 0.9099 0.9115 0.9131 0.9147 0.9162 0.9177
1.4 0.9192 0.9207 0.9222 0.9236 0.9251 0.9265 0.9279 0.9292 0.9306 0.9319
1.5 0.9332 0.9345 0.9357 0.9370 0.9382 0.9394 0.9406 0.9418 0.9429 0.9441
1.6 0.9452 0.9463 0.9474 0.9484 0.9495 0.9505 0.9515 0.9525 0.9535 0.9545
1.7 0.9554 0.9564 0.9573 0.9582 0.9591 0.9599 0.9608 0.9616 0.9625 0.9633
1.8 0.9641 0.9649 0.9656 0.9664 0.9671 0.9678 0.9686 0.9693 0.9699 0.9706
1.9 0.9713 0.9719 0.9726 0.9732 0.9738 0.9744 0.9750 0.9756 0.9761 0.9767
2.0 0.9772 0.9778 0.9783 0.9788 0.9793 0.9798 0.9803 0.9808 0.9812 0.9817
2.1 0.9821 0.9826 0.9830 0.9834 0.9838 0.9842 0.9846 0.9850 0.9854 0.9857
2.2 0.9861 0.9864 0.9868 0.9871 0.9875 0.9878 0.9881 0.9884 0.9887 0.9890
2.3 0.9893 0.9896 0.9898 0.9901 0.9904 0.9906 0.9909 0.9911 0.9913 0.9916
2.4 0.9918 0.9920 0.9922 0.9925 0.9927 0.9929 0.9931 0.9932 0.9934 0.9936
2.5 0.9938 0.9940 0.9941 0.9943 0.9945 0.9946 0.9948 0.9949 0.9951 0.9952
2.6 0.9953 0.9955 0.9956 0.9957 0.9959 0.9960 0.9961 0.9962 0.9963 0.9964
2.7 0.9965 0.9966 0.9967 0.9968 0.9969 0.9970 0.9971 0.9972 0.9973 0.9974
2.8 0.9974 0.9975 0.9976 0.9977 0.9977 0.9978 0.9979 0.9979 0.9980 0.9981
2.9 0.9981 0.9982 0.9982 0.9983 0.9984 0.9984 0.9985 0.9985 0.9986 0.9986
3.0 0.9987 0.9987 0.9987 0.9988 0.9988 0.9989 0.9989 0.9989 0.9990 0.9990
3.1 0.9990 0.9991 0.9991 0.9991 0.9992 0.9992 0.9992 0.9992 0.9993 0.9993
3.2 0.9993 0.9993 0.9994 0.9994 0.9994 0.9994 0.9994 0.9995 0.9995 0.9995
3.3 0.9995 0.9995 0.9995 0.9996 0.9996 0.9996 0.9996 0.9996 0.9996 0.9997
3.4 0.9997 0.9997 0.9997 0.9997 0.9997 0.9997 0.9997 0.9997 0.9997 0.9998
3.5 0.9998 0.9998 0.9998 0.9998 0.9998 0.9998 0.9998 0.9998 0.9998 0.9998
3.6 0.9998 0.9998 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999
3.7 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999
3.8 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999
3.9 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
4.0 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000

DET NORSKE VERITAS

Recommended Practice DNV-RP-C207, October 2010
Page 28

DET NORSKE VERITAS